[Liboil] Why every process to execute oil_init?

littertiger tiger_zsch at 163.com
Thu Aug 23 20:16:52 PDT 2007

>On Wed, Aug 22, 2007 at 09:17:00AM +0800, littertiger wrote:
>> On my arm9, oil_init takes about 20s. It seems testing implements
>> wast a lot of time. Why does every process have to do the same thing?
>> I think testing implements be executed only one time per cpu.
>I've tried to keep it under a millisecond on typical desktop
>processors, so I'm a bit surprised at 20 seconds.  It's possible
>that if floating point code is emulated in software on your CPU
>(likely), that it could take that long.
>In any case, the current strategy is insufficient for slower CPUs,
>but nobody has gotten around to fixing it.  The obvious solution is
>to cache results from one run to the next.
>For embedded projects, it makes most sense to generate the list of
>optimal implementations, and then compiling liboil only with those
>implementations.  This is sort of possible right now, just not

Thank you very much, David!

How can I generate the list of optimal implementations to avoid dynamic testing?
Are there some documents or patches?


More information about the Liboil mailing list